Mistral OCR 4: Document Recognition with Structured Output for 170 Languages (heise.de)

• Mistral OCR 4 delivers structured output with block types and confidence scores.
• Supports 170 languages in ten language groups, including rare ones.
• Self-hosting as a container possible, API price from $4/1000 pages.

"Mistral AI has released OCR 4. The model extracts not only text but also provides bounding boxes, block types, and confidence scores. It supports 170 languages in ten language groups and can be self-hosted as a container. The API costs $4 per 1000 pages, with Document AI at $5."